fix(patch): make --apply safe on descriptors, fallbacks and backups

Follow-up to the AST patcher, addressing review findings.

Correctness:
- _wrap_pytest_test captured the target with getattr, which unwraps a
  staticmethod/classmethod. Every verification of such a method raised
  TypeError, and "restoring" it left the class permanently mis-bound.
  Capture with inspect.getattr_static and re-wrap the candidate.
- Validate the spliced result on the textual fallback path too, not just
  the AST path. When the file was already unparseable the replacement is
  checked on its own, so a repair to a broken file is still possible.
- Re-indent the replacement in _textual_replace; a repair dedented to
  column 0 used to corrupt the enclosing block.
- Refuse a textual match that occurs more than once instead of patching
  the first one.
- Never overwrite an existing .heal-backup; later runs get a numbered
  suffix so the pristine copy survives.
- Preserve the file's line endings instead of round-tripping them
  through read_text/write_text.
- Take only the repaired function's own lines, and refuse a repair that
  defines anything alongside it: splicing put those in the target's
  scope, dropping them left a NameError at call time.
- Keep a missing trailing newline missing.

Reporting:
- apply_function_patch returns PatchResult(src_path, backup_path, notes)
  so "no backup written" cannot read as a backup filename, and keeping
  decorators the repair omitted is reported rather than silent.
- Warn instead of swallowing failures to restore a patched reference or
  to scan a module for aliases.

Tests: 17 new, covering descriptor re-wrap and subclass cls binding, the
changed-decorator branch, closures via outer.inner, dotted misses, backup
versioning, fallback re-indent/ambiguity/validation, CRLF round trip and
trailing newlines. Both xfail markers are fixed and removed.

The old fallback test used `this is not python` as its "unparseable"
fixture, which parses fine as an `is not` comparison, so the fallback
path had never actually run.

Author: Johin2

README.md

@@ -205,7 +205,7 @@ pytest --heal              # print proposed fix, leave files untouched
 pytest --heal-apply        # write the fix back to disk (creates a .py.heal-backup)
 pytest --heal-apply-force  # also allow modification of git-dirty files
 ```
-`--heal-apply` uses libcst for AST-faithful replacement when installed, falling back to textual replacement. It refuses to modify files with uncommitted git changes unless `--heal-apply-force` is given.
+`--heal-apply` locates the target function with Python's `ast` module and overwrites only that definition's lines, falling back to textual replacement when the file cannot be parsed or the target is not found in it. Either way the result is re-parsed before anything is written. It refuses to modify files with uncommitted git changes unless `--heal-apply-force` is given, and never overwrites an existing `.heal-backup`.
 
 ### CLI: heal a function from the command line
 ```bash

site/app/page.tsx

@@ -126,7 +126,7 @@ export default function Home() {
           <Feature
             icon={<GitBranch className="h-5 w-5" />}
             title="pytest --heal-apply"
-            body="Your failing tests trigger repair. libcst writes the accepted fix back to the file with a git-dirty guard and backup."
+            body="Your failing tests trigger repair. The accepted fix is written back to the file with a git-dirty guard and backup."
           />
           <Feature
             icon={<Activity className="h-5 w-5" />}